Can I run Llama 3.2 3B on Apple M4 Max (128GB)?
Yes: Llama 3.2 3B runs on Apple M4 Max (128GB) with F16 at roughly 48 tokens/s. Estimated memory 7.6 GiB at 8,192 tokens of context.
With F16 you can push the context to about 131,072 tokens before memory runs out (F16 KV cache).
Memory by quantisation (8,192-token context)
| Quantisation | Memory | Verdict | Est. speed |
|---|---|---|---|
| F16Unquantized reference | 7.6 GiB | Fits | ~48 tok/s |
| Q8_0Near-lossless | 4.8 GiB | Fits | ~82 tok/s |
| Q6_KVery close to F16 | 4.1 GiB | Fits | ~99 tok/s |
| Q5_K_MGreat quality/size | 3.7 GiB | Fits | ~110 tok/s |
| Q4_K_MBest default for most | 3.4 GiB | Fits | ~123 tok/s |
| IQ4_XSSmaller 4-bit (imatrix) | 3.2 GiB | Fits | ~134 tok/s |
| Q3_K_MNoticeable loss | 3.1 GiB | Fits | ~141 tok/s |
| Q2_KLast resort; big loss | 2.9 GiB | Fits | ~155 tok/s |
Apple M4 Max (128GB) offers about 128 GiB of memory (75% usable for a model) at 546 GB/s. Speed is a bandwidth-bound estimate of single-stream generation at 65% efficiency; real results vary by runtime, drivers and settings, so measure on your machine.
How this is calculated
Weights ≈ parameters × bits per weight ÷ 8. KV cache ≈ 2 × 28 layers × 8 KV heads × 128 head dim × 2 bytes per token. Decode speed ≈ efficiency × memory bandwidth ÷ bytes read per token. Learn the maths in VRAM math and memory bandwidth, or try the VRAM calculator.
Llama 3.2 3B on other hardware
Similar models on Apple M4 Max (128GB)
Running slowly or out of memory? See CUDA out of memory and the quantisation lesson.